Dense FixMatch: a simple semi-supervised learning method for pixel-wise prediction tasks

TL;DR

Dense FixMatch is a straightforward semi-supervised learning method that enhances pixel-wise prediction tasks by combining pseudo-labeling and strong data augmentation, significantly reducing the need for labeled data.

Contribution

It extends FixMatch to dense prediction tasks by introducing a matching operation on pseudo-labels, enabling effective semi-supervised learning for segmentation.

Findings

Achieves near-supervised performance with only 25% labeled data

Significantly outperforms supervised learning with limited labels

Effective on Cityscapes and Pascal VOC datasets

Abstract

We propose Dense FixMatch, a simple method for online semi-supervised learning of dense and structured prediction tasks combining pseudo-labeling and consistency regularization via strong data augmentation. We enable the application of FixMatch in semi-supervised learning problems beyond image classification by adding a matching operation on the pseudo-labels. This allows us to still use the full strength of data augmentation pipelines, including geometric transformations. We evaluate it on semi-supervised semantic segmentation on Cityscapes and Pascal VOC with different percentages of labeled data and ablate design choices and hyper-parameters. Dense FixMatch significantly improves results compared to supervised learning using only labeled data, approaching its performance with 1/4 of the labeled samples.